De-Anonymizing Social Graphs Via Node Similarity

Recently, a number of anonymization algorithms have been developed to protect the privacy of social graph data. However, in order to satisfy higher level of privacy requirements, it is sometimes impossible to maintain sufficient utility. Is it really easy to de-anonymize "lightly" anonymized social graphs? Here "light" anonymization algorithms stand for those algorithms that maintain higher data utility. To answer this question, we proposed a de-anonymization algorithm based on a node similarity measurement. Using the proposed algorithm, we evaluated the privacy risk of several "light" anonymization algorithms on real datasets.